Structurally, the simplest antibody (IgG) comprises four polypeptide chains, two heavy (H) chains and two light (L) chains inter-connected by disulphide bonds. The light chains exist in two distinct forms called kappa (.kappa.) and lambda (.lambda.). Each chain has a constant region (C) and a variable region (V). Each chain is organized into a series of domains. The light chains have two domains, corresponding to the C region and the other to the V region. The heavy chains have four domains, one corresponding to the V region and three domains (1,2 and 3) in the C region. The antibody has two arms (each arm being a Fab region), each of which has a VL and a VH region associated with each other. It is this pair of V regions (VL and VH) that differ from one antibody to another (owing to amino acid sequence variations), and which together are responsible for recognising the antigen and providing an antigen binding site (ABS). In even more detail, each V region is made up from three complementarity determining regions (CDR) separated by four framework regions (FR). The CDR's are the most variable part of the variable regions, and they perform the critical antigen binding function. The CDR regions are derived from many potential germ line sequences via a complex process involving recombination, mutation and selection.
It has been shown that the function of binding antigens can be performed by fragments of a whole antibody. Example binding fragments are (i) the Fab fragment consisting of the VL, VH, CL and CH1 domains; (ii) the Fd fragment consisting of the VH and CH1 domains; (iii) the Fv fragment consisting of the VL and VH domains of a single arm of an antibody, (iv) the dAb fragment (Ward et al., Nature 341:544-546 (1989)) which consists of a VH domain; (v) isolated CDR regions; and (vi) F(ab').sub.2 fragments, a bivalent fragment comprising two Fab fragments linked by a disulphide bridge at the hinge region.
Although the two domains of the Fv fragment are coded for by separate genes, it has proved possible to make a synthetic linker that enables them to be made as a single protein chain (known as single chain Fv (scFv)); Bird et al., Science 242:423-426 (1988); Huston et al., Proc. Natl. Acad. Sci., USA 85:5879-5883 (1988)) by recombinant methods. These scFv fragments were assembled from genes from monoclonals that had been previously isolated.
Bacteriophage have been constructed that express and display at their surface a large biologically functional binding molecule (e.g. antibody fragments, and enzymes and receptors) and which remain intact and infectious. This is described in WO 92/01047, the disclosure of which is herein incorporated by reference. Readers of the present document are urged to consult WO 92/01047 for detailed explanation of many of the procedures used in the experiments described herein. The applicants have called the structure which comprises a virus particle and a binding molecule displayed at the viral surface a "package". Where the binding molecule is an antibody, an antibody derivative or fragment, or a domain that is homologous to an immunoglobulin domain, the applicants call the package a "phage antibody" (pAb). However, except where the context demands otherwise, where the term phage antibody is used generally, it should also be interpreted as referring to any package comprising a virus particle and a biologically functional binding molecule displayed at the viral surface. pAbs have a range of applications in selecting antibody genes encoding antigen binding activities. For example, pAbs could be used for the cloning and rescue of hybridomas (Orlandi et al., Proc. Natl. Acad. Sci. USA, 86:3833-3837 (1989)), and in the screening of large combinatorial libraries (such as found in Huse et al., Science 246:1275-1281 (1989)). In particular, rounds of selection using pAbs may help in rescuing the higher affinity antibodies from the latter libraries. It may be preferable to screen small libraries derived from antigen-selected cells (Casali et al., Science 234:476-479 (1986)) to rescue the original VH/VL pairs comprising the Fv region of an antibody. The use of pAbs may also allow the construction of entirely synthetic antibodies. Furthermore, antibodies may be made which have some synthetic sequences e.g. CDRs, and some naturally derived sequences. For example, V-gene repertoires could be made in vitro by combining unrearranged V genes, with D and J segments. Libraries of pAbs could then be selected by binding to antigen, hypermutated in vitro in the antigen-binding loops or V domain framework regions, and subjected to further rounds of selection and mutagenesis.
The demonstration that a functional antigen-binding domain can be displayed on the surface of phage, has implications beyond the construction of novel antibodies. For example, if other protein domains can be displayed at the surface of a phage, phage vectors could be used to clone and select genes by the binding properties of the displayed protein. Furthermore, variants of proteins, including epitope libraries built into the surface of the protein, could be made and readily selected for binding activities. In effect, other protein architectures might serve as "nouvelle" antibodies.
The technique provides the possibility of building antibodies from first principles, taking advantage of the structural framework on which the antigen binding loops fold. In general, these loops have a limited number of conformations which generate a variety of binding sites by alternative loop combinations and by diverse side chains. Recent successes in modelling antigen binding sites augurs well for de novo design. In any case, a high resolution structure of the antigen is needed. However, the approach is attractive for making e.g. catalytic antibodies, particularly for small substrates. Here side chains or binding sites for prosthetic groups might be introduced, not only to bind selectively to the transition state of the substrate, but also to participate directly in bond making and breaking. The only question is whether the antibody architecture, specialised for binding, is the best starting point for building catalysts.
Genuine enzyme architectures, such as the triose phosphate isomerase (TIM) barrel, might be more suitable. Like antibodies, TIM enzymes also have a framework structure (a barrel of .beta.-strands and .alpha.-helices) and loops to bind substrate. Many enzymes with a diversity of catalytic properties are based on this architecture and the loops might be manipulated independently on the frameworks for design of new catalytic and binding properties. The phage selection system as provided by the present disclosure can be used to select for antigen binding activities and the CDR loops thus selected, used on either an antibody framework or a TIM barrel framework. Loops placed on a e.g. a TIM barrel framework could be further modified by mutagenesis and subjected to further selection.
One class of molecules that could be useful in this type of application are receptors. For example, a specific receptor could be displayed on the surface of the phage such that it would bind its ligand. The receptor could then be modified by, for example, in vitro mutagenesis and variants having higher binding affinity for the ligand selected. The selection may be carried out according to one or more of the formats described below.
Alternatively, the phage-receptor could be used as the basis of a rapid screening system for the binding of ligands, altered ligands, or potential drug candidates. The advantages of this system namely of simple cloning, convenient expression, standard reagents and easy handling makes the drug screening application particularly attractive. In the context of this discussion, receptor means a molecule that binds a specific, or group of specific, ligand(s). The natural receptor could be expressed on the surface of a population of cells, or it could be the extracellular domain of such a molecule (whether such a form exists naturally or not), or a soluble molecule performing a natural binding function in the plasma, or within a cell or organ.
Another possibility, is the display of an enzyme molecule or active site of an enzyme molecule on the surface of a phage (see examples 11, 12, 30, 31, 32 and 36 of WO 92/01047). Once the phage enzyme is expressed, it can be selected by affinity chromatography, for instance on columns derivatized with transition state analogues. If an enzyme with a different or modified specificity is desired, it may be possible to mutate an enzyme displayed as a fusion on bacteriophage and then select on a column derivatised with an analogue selected to have a higher affinity for an enzyme with the desired modified specificity.
Although throughout this application, the applicants discuss the possibility of screening for higher affinity variants of pAbs, they recognise that in some applications, for example low affinity chromatography (Ohlson, S. et al Anal. Biochem. 169, p204-208 (1988)), it may be desirable to isolate lower affinity variants.
pAbs also allow the selection of antibodies for improved stability. It has been noted for many antibodies, that yield and stability are improved when the antibodies are expressed at 30.degree. C. rather than 37.degree. C. If pAbs are displayed at 37.degree. C., only those which are stable will be available for affinity selection. When antibodies are to be used in vivo for therapeutic or diagnostic purposes, increased stability would extend the half-life of antibodies in circulation.
Although stability is important for all antibodies and antibody domains selected using phage, it is particularly important for the selection of Fv fragments which are formed by the non-covalent association of VH and VL fragments. Fv fragments have a tendency to dissociate and have a much reduced half-life in circulation compared to whole antibodies. Fv fragments are displayed on the surface of phage, by the association of one chain expressed as a gene III protein fusion with the complementary chain expressed as a soluble fragment. If pairs of chains have a high tendency to dissociate, they will be much less likely to be selected as pAbs. Therefore, the population will be enriched for pairs which do associate stably. Although dissociation is less of a problem with Fab fragments, selection would also occur for Fab fragments which associate stably. pAbs allow selection for stability to protease attack, only those pAbs that are not cleaved by proteases will be capable of binding their ligand and therefore populations of phage will be enriched for those displaying stable antibody domains.
The technique of displaying binding molecules on the phage surface can also be used as a primary cloning system. For example, a cDNA library can be constructed and inserted into the bacteriophage and this phage library screened for the ability to bind a ligand. The ligand/binding molecule combination could include any pair of molecules with an ability to specifically bind to one another e.g. receptor/ligand, enzyme/substrate (or analogue), nucleic acid binding protein/nucleic acid etc. If one member of the complementary pair is available, this may be a preferred way of isolating a clone for the other member of the pair.
The first functional antibody molecules to be expressed on the surface of filamentous phage were single-chain Fv's (scFv), so-called because heavy and light chain variable domains, normally on two separate proteins, are covalently joined by a flexible linker peptide. Alternative expression strategies have also been successful. Fab molecules can be displayed on phage if one of the chains (heavy or light) is fused to g3 capsid protein and the complementary chain exported to the periplasm as a soluble molecule. The two chains can be encoded on the same or on different replicons; the important point is that the two antibody chains in each fab molecule assemble post-translationally and the dimer is incorporated into the phage particle via linkage of one of the chains to g3p.
More recent cloning has been performed with `phagemid` vectors which have ca. 100-fold higher transformation efficiencies than phage DNA. These are plasmids containing the intergenic region from filamentous phages which enables single-stranded copies of the phagemid DNA to be produced, and packaged into infectious filamentous particles when cells harbouring them are infected with `helper` phages providing the phage components in trans. When phagemids contain gIII fused to an antibody gene (e.g. pHEN-1), the resulting fusion protein is displayed on the phagemid particle (Hoogenboom et al., Nucleic Acids Res. 19(15):4133-4137 (1991)). Multi-subunit proteins on the surface of filamentous phage: methodologies for displaying antibody (Fab) heavy and light chains. Efficient strategies have been developed for cloning antibody genes, a factor which becomes most important when dealing with large numbers of different antibody fragments such as repertoires.
The cloning vector fd-DOG-1 was used in early work with phage antibody repertoires in which scFv fragments were derived from spleen mRNA of mice immunised with the hapten oxazalone (Clackson et al., Nature 352:624-628 (1991)). Making antibody fragments using phage display libraries. VH and VL domains were separately amplified then linked at random via a short DNA fragment encoding the scFv linker peptide to produce a library of approxiamtely 10.sup.5 different clones. This was panned against the immunising antigen to select combinations of VH and VL which produced functional antibodies. Several binders were isolated, one in particular having an affinity not far below that of the best monoclonal antibodies produced by conventional hybridoma technology.
In a mouse, at any one time there are approximately 10.sup.7 possible H chains and 10.sup.5 possible L chains, making a total of 10.sup.12 possible VH:VL combinations when the two chains are combined at random (these figures are estimates and simply provide a rough guide to repertoire size). By these figures, the above mouse library sampled only 1 in 10.sup.7 of the possible VH:VL combinations. It is likely that good affinity antibodies were isolated in the work described in the preceeding paragraph because the spleen cells derived from an immunised donor in which B cells capable of recognising the antigen are clonally expanded and producing large quantities of Ig mRNA. The low library complexity in this experiment is partly due to the intrinsically low transformation efficiency of phage DNA compared to plasmid (or phagemid).
Marks et al. (Marks et al., By-Passing Immunization: Human Antibodies from V-Gene Libraries Displayed on Phage. J. Mol. Biol. 222:581-597 (1991)) and WO 92/01047 describe construction of an antibody repertoire from unimmunised humans cloned in the phagemid pHEN-1. This library, consisting of 3.times.10.sup.7 clones has so far yielded specific antibodies to many different antigens. These antibodies tend to have the moderate affinities expected of a primary immune response, demonstrating that usable antibodies to a range of structurally diverse antigens can indeed be isolated from a single resource.
New binders can be created from clones isolated from phage antibody libraries using a procedure called `chain-shuffling`. In this process one of the two chains is fixed and the other varied. For example, by fixing the heavy chain from the highest affinity mouse anti-OX phage antibody and recloning the repertoire of light chains alongside it, libraries of 4.times.10.sup.7 were constructed. Several new OX-binders were isolated, and the majority of these had light chains that were distinct from those first isolated and considerably more diverse. These observations reflect the fact that a small library is sufficient to tap the available diversity when only one chain is varied, a useful procedure if the original library was not sufficiently large to contain the available diversity.
The size of the library is of critical importance. This is especially true when attempting to isolate antibodies from a naive human repertoire, but is equally relevant to isolation of the highest affinity antibodies from an immunised source.
It is clear that while phage display is an exceptionally powerful tool for cloning and selecting antibody genes, we are tapping only the tiniest fraction of the potential diversity using existing technology. Transformation efficiencies place the greatest limitation on library size with 10.sup.9 being about the limit using current methods. Rough calculations suggest that this is several orders of magnitude below the target efficiency; more rigourous analysis confirms it.
Perelson and Oster have given theoretical consideration to the relationship between size of the immune repertoire and the likelihood of generating an antibody capable recognising a given epitope with greater than a certain threshold affinity, K. The relationship is described by the equation: EQU P=e.sup.-N (pK!)
Where P=probability that an epitope is not recognised with an affinity above the threshold value K by any antibody in the repertoire,
N=number of different antibodies in the repertoire, and
pK!=probability that an individual antibody recognises a random epitope with an affinity above the threshold value K
In this analysis pK! is inversely proportional to affinity, although an algorithm describing this relationship precisely has not been deduced. Despite this, it is apparent that the higher the affinity of the antibody, the lower its pK! and the larger the repertoire needs to be to achieve a reasonable probability of isolating that antibody. The other important feature is that the function is exponential; as shown in FIG. 1, a small change in library size can have either a negligible or a dramatic effect on the probability of isolating an antibody with a given pK! value, depending upon what point on the curve is given by the library size.
WO 92/01047 and WO 92/20791 (also incorporated herein by reference) describe how the limitations of transformation efficiency (and therefore the upper limit on library size) can be overcome by use of other methods for introducing DNA into cells, such as infection. In one configuration, heavy and light chain genes are cloned separately on two different replicons, at least one of which is capable of being incorporated into a filamentous particle. Infectious particles carrying one chain are infected into cells harbouring the complementary chain; infection frequencies of &gt;90% can be readily achieved. Heavy and light chains are then able to associate post-translationally in the periplasm and the combination displayed on the surface of the filamentous particle by virtue of one or both chains being connected to g3p. For example, a library of 10.sup.7 heavy chains is cloned as an unfused population in a phagemid, and 10.sup.7 light chains are cloned as g3 fusions in fd-DOG-1. Both populations are then expanded by growth such that there are 10.sup.7 of each heavy chain-containing cell and 10.sup.7 copies of each light chain phage. By allowing the phage to infect the cells, 10.sup.7 .times.10.sup.7 =10.sup.14 unique combinations can be created, because there are 10.sup.7 cells carrying the same heavy chain which can each be infected by 10.sup.7 phage carrying different light chains. When this is repeated for each different heavy chain clone then one ends up with up to 10.sup.14 different heavy/light combinations in different cells. This strategy is outlined in FIG. 2, which shows the heavy chain cloned as g3 fusions on phage and the light chains expressed as soluble fragments from a phagemid. Clearly, the reverse combination, light chains on phage, heavy chain on phagemid, is also tenable.
In the configuration shown in FIG. 2, fd-DOG `rescues` the phagemid so that both phage and phagemid DNA is packaged into filamentous particles, and both types will have paired heavy and light chains on their surface, despite having the genetic information for only one of them. For a given antigen or epitope, the vast majority of the heavy and light chain pairings will be non-functional (i.e. will not bind that antigen or epitope), so that selection on antigen will have the effect of vastly reducing the complexity of the heavy and light chain populations. After the first round of selection the clones are re-assorted, for example by infecting fresh host cells and selecting for both replicons. After several rounds of antigen selection and recovery of the two replicons, the considerably reduced heavy and light chain populations can be cloned onto the same replicon and analysed by conventional means. Selection from the, say, 10.sup.14 combinations produces a population of phages displaying a particular combination of H and L chains having the desired specificity. The phages selected however, will only contain DNA encoding one partner of the paired H and L chains. Selection for the two replicons may be as follows. Vectors of the H chain library may encode tetracycline resistance, with vectors of the L chain library encoding ampicillin resistance. The sample elute containing the population is divided into two portions. A first portion is grown on e.g. tetracycline plates to select those bacteriophage containing DNA encoding H chains which are involved in the desired antigen binding. A second portion is grown on e.g. ampicillin plates to select those bacteriophage containing phagemid DNA encoding L chains which are involved in the desired antigen binding. A set of colonies from individually isolated clones e.g. from the tetracycline plates are then used to infect specific colonies e.g. from the ampicillin plates. This results in bacteriophage expressing specific combinations of H and L chains which can then be assayed for antigen binding.
One technical problem with the use of separate replicons for VL and VH chains is so-called `interference` between filamentous phage origins of replication carried on different replicons as a result of competition for the same replication machinery.
Procedures have been described which work on the principle of first reducing the complexity of a repertoire then recloning one or both chains of the reduced population (WO 092/20791). The present invention provides a different approach.